The Yield Anesthetic: AI's Debt Supercycle and the Risk the Market Has Not Priced

How more than half a trillion dollars of AI-related issuance is being absorbed at the tightest spreads in a generation, why that calm is a function of all-in yield rather than credit discipline, and what the migration of risk into private credit means for anyone raising or deploying capital in technology.

There is a comfortable story being told in credit markets right now. The largest technology companies are issuing debt at a historic pace to fund artificial intelligence infrastructure, and the market is absorbing every dollar without complaint. Spreads sit near their tightest levels since the late 1990s. Credit volatility is close to record lows. The conclusion most observers draw is that investors have looked at these balance sheets and decided the risk is negligible.

That conclusion is half right, and the wrong half is the half that matters.

Investors are not buying these bonds because they have underwritten the credit. They are buying them because the all-in yield is compelling. With government bond yields elevated, an investment-grade corporate bond paying close to 6 percent clears the hurdle on absolute return alone, almost regardless of what the spread is doing. The spread, the actual compensation for credit risk, has been anesthetized by the level of the underlying rate. That is a very different thing from a market that has examined the AI buildout and priced it correctly. It is a market that has not yet been required to ask the question.

The Supply Math Is No Longer Incremental

The scale here is not a continuation of trend. It is a structural break. The five largest hyperscalers are on track to spend in the region of $600 billion on capital expenditure in 2026, with the most aggressive estimates running toward $770 billion, the substantial majority of it tied to AI infrastructure. For several of these issuers, capex is approaching the entirety of operating cash flow, against a historical norm closer to 40 percent. Capex-to-revenue ratios that would once have been considered untenable, north of 45 percent at multiple firms, are now the operating assumption.

Internal cash can no longer carry that load, so the funding shifts to the capital markets. Hyperscalers raised on the order of $108 billion to $121 billion in debt in 2025, more than four times their average annual issuance over the prior five years. Street estimates for AI-related investment-grade supply in 2026 cluster around $300 billion, with longer-horizon projections from major banks pointing to $1.5 trillion of investment-grade issuance over the next five years and total AI-related capital needs measured in the multiple trillions through the end of the decade.

The reassuring counterpoint is that the public investment-grade market is enormous, and these issuers remain a small share of it. The five largest names represent close to 20 percent of the equity market but only around 3.5 percent of public investment-grade debt. There is, on paper, considerable room to absorb more. The constraint is not the size of the market in aggregate. It is concentration. Index-tracking buyers and portfolios with single-issuer or single-sector limits will reach saturation against the same handful of names well before the broad market does. When that happens, the marginal buyer demands a concession, and the concession reprices the curve.

The Risk Is Migrating to Where the Disclosure Is Weakest

The more important development is structural, and it is largely invisible in the headline spread. A growing share of AI infrastructure is being financed off the corporate balance sheet entirely, through special-purpose vehicles, joint ventures, and private credit. The hyperscaler takes a minority stake, signs a long-term lease or capacity offtake, and sometimes provides a guarantee. The debt is serviced by lease cash flows and held by private credit funds and insurers. The economic substance is borrowing. The accounting presentation is an operating expense. The Bank for International Settlements has a precise term for this: shadow borrowing.

This is where the comfortable story breaks down. Public investment-grade bonds from AA and AAA issuers are the most visible and least concerning part of the picture. The financing that warrants scrutiny is the part that has migrated into structures with customized terms, estimated rather than marked valuations, and limited liquidity. Pricing tells the story. Where a hyperscaler's senior unsecured debt trades at a modest spread, the private financing of the same company's data center assets can price at roughly double that, even with asset backing and contractual support. The market is charging materially more for the same credit once it sits inside an opaque structure. That differential is not noise. It is the risk premium the public spread is not showing you.

Layer onto this the asset-life problem. Much of this debt is long-dated, funding assets whose useful economic life is genuinely uncertain. If models become dramatically more efficient, or a hardware generation renders current capacity uncompetitive, the buildings and the chips inside them could be impaired well before the debt that financed them matures. The fiber-optic overbuild of the early 2000s is the obvious parallel: the long-term utility of the asset was eventually vindicated, but the issuers who financed the overcapacity did not survive the interval.

The Monetization Gap Is the Variable to Watch

None of this is a problem so long as the revenue arrives on schedule. It is the gap between spending and earning that converts a financing question into a credit question. Annual hyperscaler AI capex is running near $400 billion against realized AI revenue estimated closer to $100 billion. That gap can be carried for a long time by companies with the cash flows these firms generate. It cannot be carried indefinitely, and it does not need to close suddenly to matter. It only needs to disappoint relative to expectations at a moment when spreads offer no cushion. From 25-year tights, there is very little risk premium left to absorb a shock.

What This Means for Capital Formation, Not Just for Bondholders

For the firms and investors Chatsworth advises, the bond-market debate is the wrong frame. The variable that matters is the cost of capital, and the trajectory of these flows points in one direction.

First, the duration supply from AI issuance is large enough to steepen the long end of the curve and to crowd out other investment-grade issuers, financials in particular. That is a system-wide repricing, not a technology-sector event. The cost of long-dated capital for every issuer is being shaped at the margin by five companies' infrastructure ambitions.

Second, a higher and more volatile cost of capital flows directly into technology valuations. Discount rates do the heavy lifting in any growth-asset valuation, and a credit market that reprices will reset both the multiples buyers are willing to pay and the windows in which sellers can transact. The current calm is, in this sense, an asset for anyone contemplating a financing or a sale, and a depreciating one.

Third, the bifurcation is the opportunity and the trap. The market is already separating committed, transparent, investment-grade hyperscaler credit from sub-investment-grade neocloud and GPU-as-a-service tenants whose underwriting rests on far less. The same is now visible in private credit more broadly, where a string of defaults and at least one large fund pausing redemptions has reminded allocators that the tide has not, until recently, gone out on the asset class. Disciplined capital will be rewarded for distinguishing the two. Capital that treats all AI-adjacent exposure as a single trade will not.

The Question That Actually Matters

The headline question, whether AI debt will strain credit markets, is answerable but not yet useful, because the strain has not been tested. The better question is narrower and more demanding: how long can absolute yield continue to substitute for credit discipline, and what is the state of the most opaque corners of this capital structure on the day it stops?

Spreads at generational tights are not a verdict that the risk is small. They are a statement that the risk has not yet been asked to clear a market on its own terms. When it is, the repricing will not be confined to the bonds. It will move through the cost of capital, into valuations, and into the timing of every transaction that depends on them. The firms that have read both the balance sheet and the structure beneath it will be the ones positioned for it.

Chatsworth Securities LLC is a dually registered broker-dealer and SEC-registered investment adviser. This material is provided for informational purposes only. It does not constitute investment, legal, or tax advice, a recommendation, or an offer or solicitation to buy or sell any security or financial instrument. Figures cited reflect third-party estimates available as of publication and are subject to revision. Forward-looking statements are inherently uncertain, and past performance is not indicative of future results.